Guides
AI Guides.
Practical guides to deploying AI — for funds, enterprise and SMB, plus decisions and comparisons. Each links to the glossary and a concrete offering.
For PE funds
AI/IT due diligence: how to assess a company's "AI" before you sign
AI due diligence separates real capability from a wrapper on someone else's model and from data debt. A time-boxed risk review that fits into a board deck.
The first 100 days: a value-creation map for AI in a portfolio company
A plan for the first 100 days after close: sequence automations by EBITDA impact and risk, start with guardrails, and decide — centralized or per company.
Hidden AI risks in a deal: data debt, vendor dependence, the AI Act
Some AI liabilities survive the close: data provenance, vendor lock-in, AI Act exposure. Framed as deal economics, not a technical topic.
For enterprise
Agentic architecture that passes audit: HITL, guardrails, MCP, audit trail
An agentic architecture ready for audit has named control points: a human in the loop, guardrails, controlled access through MCP, and a full audit trail of every decision.
Why AI pilots don't scale — and how to fix it with architecture
AI pilots don't die on the model — they die on data access, evaluation, ownership, audit, and cost. Solve those blockers first, then pick the tools.
Governing AI agents before they spread
Governing AI agents is an operating model: model routing, access control, logging, and evaluation gates. Put in early, before agents multiply, it's a lever, not a brake.
For SMB
AI and GDPR in a small company: customer data and simple safeguards
GDPR with AI doesn't take a big budget. The key is knowing which data not to paste, plus a few simple safeguards you can put in place right away.
Small pilot, hard proof, then scale: AI in SMEs without an in-house team
The cheapest route to AI in a small company: a narrow pilot with one success metric, clear stop conditions, and scale only after hard proof.
Where to start with AI in an SME: the first process that pays for itself
Pick your first process by three criteria: volume, repeatability and the cost of getting it wrong. Start with proposals, support or documents — not everything at once.
A workflow that pays for itself: how to calculate the return on automation before you deploy
You calculate the return with a simple formula: time saved times the rate, minus the cost of running it. Do it on a napkin before you sign off on a deployment.
Decisions & comparisons
Assistant, workflow, or AI agent — how to choose the level of autonomy
An assistant waits for a command, a workflow follows fixed steps, an agent plans its own loop. You match the level of autonomy to the cost of a mistake, under human supervision.
AI audit vs proof of concept: what to buy first
An AI audit assesses what you have and where the leverage is. A PoC builds one case to test it. The audit picks the target, the PoC tests it — usually in that order.
AI system security: prompt injection, data leaks, and how to defend
An AI system has its own attack surface: prompt injection, data leaks through context, tool abuse. The defense is layers, not a single filter.
Build vs buy: build your own agent or buy off the shelf
An off-the-shelf agent goes live in days and hands maintenance to the vendor; a custom agent gives control over data and logic, at the cost of time. It depends on the process.
What is agentic AI — and how it differs from a chatbot and from automation
Agentic AI is a system that plans its own next steps, uses tools, and checks the result. A chatbot answers a question; rigid automation follows a fixed path.
The EU AI Act for businesses: how to organize your knowledge and data to be ready
The EU AI Act means obligations that scale with the level of risk: transparency, documentation, human oversight, and data governance. Organized knowledge makes compliance easier.
AI hallucinations: why a model makes things up and how to prevent it
A hallucination is a confident-sounding, false model answer. It doesn't disappear entirely — you limit it: with RAG, rules, evaluation, and human oversight.
What AI in production really costs: tokens, inference, maintenance
The per-token price is a fraction of the bill. Real cost is driven by context length, number of calls, retrieval, and maintenance. What matters is the cost per task completed.
How to tell whether an AI agent works: evaluation, not impression
Judging an agent "by feel" confuses a good demo with a working system. A fixed set of cases and the right metrics give you proof instead of an impression.
How to choose a language model: the right one for the task, not the "best"
There's no single best model — only the one that fits the task. What matters is cost, context window, privacy, hosting, and latency — not a leaderboard ranking.
MCP and agent integrations: how to connect AI to your own systems
MCP is a shared protocol through which an agent connects to tools, data, and APIs. It decides what the agent can reach — the limits and approvals stay on the human's side.
RAG or fine-tuning — when to use which (and when to use both)
RAG adds knowledge to a model by searching your documents; fine-tuning changes the model's behavior itself. Most of the time you start with RAG.